1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
4 * Copyright (C) 2004-2006 Red Hat, Inc. All rights reserved.
5 */
6
7#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
8
9#include <linux/spinlock.h>
10#include <linux/completion.h>
11#include <linux/buffer_head.h>
12#include <linux/kthread.h>
13#include <linux/crc32.h>
14#include <linux/gfs2_ondisk.h>
15#include <linux/delay.h>
16#include <linux/uaccess.h>
17
18#include "gfs2.h"
19#include "incore.h"
20#include "glock.h"
21#include "glops.h"
22#include "log.h"
23#include "lops.h"
24#include "recovery.h"
25#include "rgrp.h"
26#include "super.h"
27#include "util.h"
28
29struct kmem_cache *gfs2_glock_cachep __read_mostly;
30struct kmem_cache *gfs2_glock_aspace_cachep __read_mostly;
31struct kmem_cache *gfs2_inode_cachep __read_mostly;
32struct kmem_cache *gfs2_bufdata_cachep __read_mostly;
33struct kmem_cache *gfs2_rgrpd_cachep __read_mostly;
34struct kmem_cache *gfs2_quotad_cachep __read_mostly;
35struct kmem_cache *gfs2_qadata_cachep __read_mostly;
36struct kmem_cache *gfs2_trans_cachep __read_mostly;
37mempool_t *gfs2_page_pool __read_mostly;
38
39void gfs2_assert_i(struct gfs2_sbd *sdp)
40{
41 fs_emerg(sdp, "fatal assertion failed\n");
42}
43
44/**
45 * check_journal_clean - Make sure a journal is clean for a spectator mount
46 * @sdp: The GFS2 superblock
47 * @jd: The journal descriptor
48 * @verbose: Show more prints in the log
49 *
50 * Returns: 0 if the journal is clean or locked, else an error
51 */
52int check_journal_clean(struct gfs2_sbd *sdp, struct gfs2_jdesc *jd,
53 bool verbose)
54{
55 int error;
56 struct gfs2_holder j_gh;
57 struct gfs2_log_header_host head;
58 struct gfs2_inode *ip;
59
60 ip = GFS2_I(inode: jd->jd_inode);
61 error = gfs2_glock_nq_init(gl: ip->i_gl, LM_ST_SHARED, LM_FLAG_NOEXP |
62 GL_EXACT | GL_NOCACHE, gh: &j_gh);
63 if (error) {
64 if (verbose)
65 fs_err(sdp, "Error %d locking journal for spectator "
66 "mount.\n", error);
67 return -EPERM;
68 }
69 error = gfs2_jdesc_check(jd);
70 if (error) {
71 if (verbose)
72 fs_err(sdp, "Error checking journal for spectator "
73 "mount.\n");
74 goto out_unlock;
75 }
76 error = gfs2_find_jhead(jd, head: &head, keep_cache: false);
77 if (error) {
78 if (verbose)
79 fs_err(sdp, "Error parsing journal for spectator "
80 "mount.\n");
81 goto out_unlock;
82 }
83 if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) {
84 error = -EPERM;
85 if (verbose)
86 fs_err(sdp, "jid=%u: Journal is dirty, so the first "
87 "mounter must not be a spectator.\n",
88 jd->jd_jid);
89 }
90
91out_unlock:
92 gfs2_glock_dq_uninit(gh: &j_gh);
93 return error;
94}
95
96/**
97 * gfs2_freeze_lock_shared - hold the freeze glock
98 * @sdp: the superblock
99 */
100int gfs2_freeze_lock_shared(struct gfs2_sbd *sdp)
101{
102 int error;
103
104 error = gfs2_glock_nq_init(gl: sdp->sd_freeze_gl, LM_ST_SHARED,
105 LM_FLAG_NOEXP | GL_EXACT,
106 gh: &sdp->sd_freeze_gh);
107 if (error)
108 fs_err(sdp, "can't lock the freeze glock: %d\n", error);
109 return error;
110}
111
112void gfs2_freeze_unlock(struct gfs2_holder *freeze_gh)
113{
114 if (gfs2_holder_initialized(gh: freeze_gh))
115 gfs2_glock_dq_uninit(gh: freeze_gh);
116}
117
118static void signal_our_withdraw(struct gfs2_sbd *sdp)
119{
120 struct gfs2_glock *live_gl = sdp->sd_live_gh.gh_gl;
121 struct inode *inode;
122 struct gfs2_inode *ip;
123 struct gfs2_glock *i_gl;
124 u64 no_formal_ino;
125 int ret = 0;
126 int tries;
127
128 if (test_bit(SDF_NORECOVERY, &sdp->sd_flags) || !sdp->sd_jdesc)
129 return;
130
131 gfs2_ail_drain(sdp); /* frees all transactions */
132 inode = sdp->sd_jdesc->jd_inode;
133 ip = GFS2_I(inode);
134 i_gl = ip->i_gl;
135 no_formal_ino = ip->i_no_formal_ino;
136
137 /* Prevent any glock dq until withdraw recovery is complete */
138 set_bit(nr: SDF_WITHDRAW_RECOVERY, addr: &sdp->sd_flags);
139 /*
140 * Don't tell dlm we're bailing until we have no more buffers in the
141 * wind. If journal had an IO error, the log code should just purge
142 * the outstanding buffers rather than submitting new IO. Making the
143 * file system read-only will flush the journal, etc.
144 *
145 * During a normal unmount, gfs2_make_fs_ro calls gfs2_log_shutdown
146 * which clears SDF_JOURNAL_LIVE. In a withdraw, we must not write
147 * any UNMOUNT log header, so we can't call gfs2_log_shutdown, and
148 * therefore we need to clear SDF_JOURNAL_LIVE manually.
149 */
150 clear_bit(nr: SDF_JOURNAL_LIVE, addr: &sdp->sd_flags);
151 if (!sb_rdonly(sb: sdp->sd_vfs)) {
152 bool locked = mutex_trylock(lock: &sdp->sd_freeze_mutex);
153
154 wake_up(&sdp->sd_logd_waitq);
155 wake_up(&sdp->sd_quota_wait);
156
157 wait_event_timeout(sdp->sd_log_waitq,
158 gfs2_log_is_empty(sdp),
159 HZ * 5);
160
161 sdp->sd_vfs->s_flags |= SB_RDONLY;
162
163 if (locked)
164 mutex_unlock(lock: &sdp->sd_freeze_mutex);
165
166 /*
167 * Dequeue any pending non-system glock holders that can no
168 * longer be granted because the file system is withdrawn.
169 */
170 gfs2_gl_dq_holders(sdp);
171 }
172
173 if (sdp->sd_lockstruct.ls_ops->lm_lock == NULL) { /* lock_nolock */
174 if (!ret)
175 ret = -EIO;
176 clear_bit(nr: SDF_WITHDRAW_RECOVERY, addr: &sdp->sd_flags);
177 goto skip_recovery;
178 }
179 /*
180 * Drop the glock for our journal so another node can recover it.
181 */
182 if (gfs2_holder_initialized(gh: &sdp->sd_journal_gh)) {
183 gfs2_glock_dq_wait(gh: &sdp->sd_journal_gh);
184 gfs2_holder_uninit(gh: &sdp->sd_journal_gh);
185 }
186 sdp->sd_jinode_gh.gh_flags |= GL_NOCACHE;
187 gfs2_glock_dq(gh: &sdp->sd_jinode_gh);
188 gfs2_thaw_freeze_initiator(sb: sdp->sd_vfs);
189 wait_on_bit(word: &i_gl->gl_flags, bit: GLF_DEMOTE, TASK_UNINTERRUPTIBLE);
190
191 /*
192 * holder_uninit to force glock_put, to force dlm to let go
193 */
194 gfs2_holder_uninit(gh: &sdp->sd_jinode_gh);
195
196 /*
197 * Note: We need to be careful here:
198 * Our iput of jd_inode will evict it. The evict will dequeue its
199 * glock, but the glock dq will wait for the withdraw unless we have
200 * exception code in glock_dq.
201 */
202 iput(inode);
203 sdp->sd_jdesc->jd_inode = NULL;
204 /*
205 * Wait until the journal inode's glock is freed. This allows try locks
206 * on other nodes to be successful, otherwise we remain the owner of
207 * the glock as far as dlm is concerned.
208 */
209 if (i_gl->gl_ops->go_free) {
210 set_bit(nr: GLF_FREEING, addr: &i_gl->gl_flags);
211 wait_on_bit(word: &i_gl->gl_flags, bit: GLF_FREEING, TASK_UNINTERRUPTIBLE);
212 }
213
214 /*
215 * Dequeue the "live" glock, but keep a reference so it's never freed.
216 */
217 gfs2_glock_hold(gl: live_gl);
218 gfs2_glock_dq_wait(gh: &sdp->sd_live_gh);
219 /*
220 * We enqueue the "live" glock in EX so that all other nodes
221 * get a demote request and act on it. We don't really want the
222 * lock in EX, so we send a "try" lock with 1CB to produce a callback.
223 */
224 fs_warn(sdp, "Requesting recovery of jid %d.\n",
225 sdp->sd_lockstruct.ls_jid);
226 gfs2_holder_reinit(LM_ST_EXCLUSIVE,
227 LM_FLAG_TRY_1CB | LM_FLAG_NOEXP | GL_NOPID,
228 gh: &sdp->sd_live_gh);
229 msleep(GL_GLOCK_MAX_HOLD);
230 /*
231 * This will likely fail in a cluster, but succeed standalone:
232 */
233 ret = gfs2_glock_nq(gh: &sdp->sd_live_gh);
234
235 /*
236 * If we actually got the "live" lock in EX mode, there are no other
237 * nodes available to replay our journal. So we try to replay it
238 * ourselves. We hold the "live" glock to prevent other mounters
239 * during recovery, then just dequeue it and reacquire it in our
240 * normal SH mode. Just in case the problem that caused us to
241 * withdraw prevents us from recovering our journal (e.g. io errors
242 * and such) we still check if the journal is clean before proceeding
243 * but we may wait forever until another mounter does the recovery.
244 */
245 if (ret == 0) {
246 fs_warn(sdp, "No other mounters found. Trying to recover our "
247 "own journal jid %d.\n", sdp->sd_lockstruct.ls_jid);
248 if (gfs2_recover_journal(gfs2_jd: sdp->sd_jdesc, wait: 1))
249 fs_warn(sdp, "Unable to recover our journal jid %d.\n",
250 sdp->sd_lockstruct.ls_jid);
251 gfs2_glock_dq_wait(gh: &sdp->sd_live_gh);
252 gfs2_holder_reinit(LM_ST_SHARED,
253 LM_FLAG_NOEXP | GL_EXACT | GL_NOPID,
254 gh: &sdp->sd_live_gh);
255 gfs2_glock_nq(gh: &sdp->sd_live_gh);
256 }
257
258 gfs2_glock_queue_put(gl: live_gl); /* drop extra reference we acquired */
259 clear_bit(nr: SDF_WITHDRAW_RECOVERY, addr: &sdp->sd_flags);
260
261 /*
262 * At this point our journal is evicted, so we need to get a new inode
263 * for it. Once done, we need to call gfs2_find_jhead which
264 * calls gfs2_map_journal_extents to map it for us again.
265 *
266 * Note that we don't really want it to look up a FREE block. The
267 * GFS2_BLKST_FREE simply overrides a block check in gfs2_inode_lookup
268 * which would otherwise fail because it requires grabbing an rgrp
269 * glock, which would fail with -EIO because we're withdrawing.
270 */
271 inode = gfs2_inode_lookup(sb: sdp->sd_vfs, DT_UNKNOWN,
272 no_addr: sdp->sd_jdesc->jd_no_addr, no_formal_ino,
273 GFS2_BLKST_FREE);
274 if (IS_ERR(ptr: inode)) {
275 fs_warn(sdp, "Reprocessing of jid %d failed with %ld.\n",
276 sdp->sd_lockstruct.ls_jid, PTR_ERR(inode));
277 goto skip_recovery;
278 }
279 sdp->sd_jdesc->jd_inode = inode;
280 d_mark_dontcache(inode);
281
282 /*
283 * Now wait until recovery is complete.
284 */
285 for (tries = 0; tries < 10; tries++) {
286 ret = check_journal_clean(sdp, jd: sdp->sd_jdesc, verbose: false);
287 if (!ret)
288 break;
289 msleep(HZ);
290 fs_warn(sdp, "Waiting for journal recovery jid %d.\n",
291 sdp->sd_lockstruct.ls_jid);
292 }
293skip_recovery:
294 if (!ret)
295 fs_warn(sdp, "Journal recovery complete for jid %d.\n",
296 sdp->sd_lockstruct.ls_jid);
297 else
298 fs_warn(sdp, "Journal recovery skipped for jid %d until next "
299 "mount.\n", sdp->sd_lockstruct.ls_jid);
300 fs_warn(sdp, "Glock dequeues delayed: %lu\n", sdp->sd_glock_dqs_held);
301 sdp->sd_glock_dqs_held = 0;
302 wake_up_bit(word: &sdp->sd_flags, bit: SDF_WITHDRAW_RECOVERY);
303}
304
305void gfs2_lm(struct gfs2_sbd *sdp, const char *fmt, ...)
306{
307 struct va_format vaf;
308 va_list args;
309
310 if (sdp->sd_args.ar_errors == GFS2_ERRORS_WITHDRAW &&
311 test_bit(SDF_WITHDRAWN, &sdp->sd_flags))
312 return;
313
314 va_start(args, fmt);
315 vaf.fmt = fmt;
316 vaf.va = &args;
317 fs_err(sdp, "%pV", &vaf);
318 va_end(args);
319}
320
321int gfs2_withdraw(struct gfs2_sbd *sdp)
322{
323 struct lm_lockstruct *ls = &sdp->sd_lockstruct;
324 const struct lm_lockops *lm = ls->ls_ops;
325
326 if (sdp->sd_args.ar_errors == GFS2_ERRORS_WITHDRAW) {
327 unsigned long old = READ_ONCE(sdp->sd_flags), new;
328
329 do {
330 if (old & BIT(SDF_WITHDRAWN)) {
331 wait_on_bit(word: &sdp->sd_flags,
332 bit: SDF_WITHDRAW_IN_PROG,
333 TASK_UNINTERRUPTIBLE);
334 return -1;
335 }
336 new = old | BIT(SDF_WITHDRAWN) | BIT(SDF_WITHDRAW_IN_PROG);
337 } while (unlikely(!try_cmpxchg(&sdp->sd_flags, &old, new)));
338
339 fs_err(sdp, "about to withdraw this file system\n");
340 BUG_ON(sdp->sd_args.ar_debug);
341
342 signal_our_withdraw(sdp);
343
344 kobject_uevent(kobj: &sdp->sd_kobj, action: KOBJ_OFFLINE);
345
346 if (!strcmp(sdp->sd_lockstruct.ls_ops->lm_proto_name, "lock_dlm"))
347 wait_for_completion(&sdp->sd_wdack);
348
349 if (lm->lm_unmount) {
350 fs_err(sdp, "telling LM to unmount\n");
351 lm->lm_unmount(sdp);
352 }
353 set_bit(nr: SDF_SKIP_DLM_UNLOCK, addr: &sdp->sd_flags);
354 fs_err(sdp, "File system withdrawn\n");
355 dump_stack();
356 clear_bit(nr: SDF_WITHDRAW_IN_PROG, addr: &sdp->sd_flags);
357 smp_mb__after_atomic();
358 wake_up_bit(word: &sdp->sd_flags, bit: SDF_WITHDRAW_IN_PROG);
359 }
360
361 if (sdp->sd_args.ar_errors == GFS2_ERRORS_PANIC)
362 panic(fmt: "GFS2: fsid=%s: panic requested\n", sdp->sd_fsname);
363
364 return -1;
365}
366
367/*
368 * gfs2_assert_withdraw_i - Cause the machine to withdraw if @assertion is false
369 */
370
371void gfs2_assert_withdraw_i(struct gfs2_sbd *sdp, char *assertion,
372 const char *function, char *file, unsigned int line,
373 bool delayed)
374{
375 if (gfs2_withdrawing_or_withdrawn(sdp))
376 return;
377
378 fs_err(sdp,
379 "fatal: assertion \"%s\" failed\n"
380 " function = %s, file = %s, line = %u\n",
381 assertion, function, file, line);
382
383 /*
384 * If errors=panic was specified on mount, it won't help to delay the
385 * withdraw.
386 */
387 if (sdp->sd_args.ar_errors == GFS2_ERRORS_PANIC)
388 delayed = false;
389
390 if (delayed)
391 gfs2_withdraw_delayed(sdp);
392 else
393 gfs2_withdraw(sdp);
394 dump_stack();
395}
396
397/*
398 * gfs2_assert_warn_i - Print a message to the console if @assertion is false
399 */
400
401void gfs2_assert_warn_i(struct gfs2_sbd *sdp, char *assertion,
402 const char *function, char *file, unsigned int line)
403{
404 if (time_before(jiffies,
405 sdp->sd_last_warning +
406 gfs2_tune_get(sdp, gt_complain_secs) * HZ))
407 return;
408
409 if (sdp->sd_args.ar_errors == GFS2_ERRORS_WITHDRAW)
410 fs_warn(sdp, "warning: assertion \"%s\" failed at function = %s, file = %s, line = %u\n",
411 assertion, function, file, line);
412
413 if (sdp->sd_args.ar_debug)
414 BUG();
415 else
416 dump_stack();
417
418 if (sdp->sd_args.ar_errors == GFS2_ERRORS_PANIC)
419 panic(fmt: "GFS2: fsid=%s: warning: assertion \"%s\" failed\n"
420 "GFS2: fsid=%s: function = %s, file = %s, line = %u\n",
421 sdp->sd_fsname, assertion,
422 sdp->sd_fsname, function, file, line);
423
424 sdp->sd_last_warning = jiffies;
425}
426
427/*
428 * gfs2_consist_i - Flag a filesystem consistency error and withdraw
429 */
430
431void gfs2_consist_i(struct gfs2_sbd *sdp, const char *function,
432 char *file, unsigned int line)
433{
434 gfs2_lm(sdp,
435 fmt: "fatal: filesystem consistency error - function = %s, file = %s, line = %u\n",
436 function, file, line);
437 gfs2_withdraw(sdp);
438}
439
440/*
441 * gfs2_consist_inode_i - Flag an inode consistency error and withdraw
442 */
443
444void gfs2_consist_inode_i(struct gfs2_inode *ip,
445 const char *function, char *file, unsigned int line)
446{
447 struct gfs2_sbd *sdp = GFS2_SB(inode: &ip->i_inode);
448
449 gfs2_lm(sdp,
450 fmt: "fatal: filesystem consistency error\n"
451 " inode = %llu %llu\n"
452 " function = %s, file = %s, line = %u\n",
453 (unsigned long long)ip->i_no_formal_ino,
454 (unsigned long long)ip->i_no_addr,
455 function, file, line);
456 gfs2_dump_glock(NULL, gl: ip->i_gl, fsid: 1);
457 gfs2_withdraw(sdp);
458}
459
460/*
461 * gfs2_consist_rgrpd_i - Flag a RG consistency error and withdraw
462 */
463
464void gfs2_consist_rgrpd_i(struct gfs2_rgrpd *rgd,
465 const char *function, char *file, unsigned int line)
466{
467 struct gfs2_sbd *sdp = rgd->rd_sbd;
468 char fs_id_buf[sizeof(sdp->sd_fsname) + 7];
469
470 sprintf(buf: fs_id_buf, fmt: "fsid=%s: ", sdp->sd_fsname);
471 gfs2_rgrp_dump(NULL, rgd, fs_id_buf);
472 gfs2_lm(sdp,
473 fmt: "fatal: filesystem consistency error\n"
474 " RG = %llu\n"
475 " function = %s, file = %s, line = %u\n",
476 (unsigned long long)rgd->rd_addr,
477 function, file, line);
478 gfs2_dump_glock(NULL, gl: rgd->rd_gl, fsid: 1);
479 gfs2_withdraw(sdp);
480}
481
482/*
483 * gfs2_meta_check_ii - Flag a magic number consistency error and withdraw
484 * Returns: -1 if this call withdrew the machine,
485 * -2 if it was already withdrawn
486 */
487
488int gfs2_meta_check_ii(struct gfs2_sbd *sdp, struct buffer_head *bh,
489 const char *type, const char *function, char *file,
490 unsigned int line)
491{
492 int me;
493
494 gfs2_lm(sdp,
495 fmt: "fatal: invalid metadata block\n"
496 " bh = %llu (%s)\n"
497 " function = %s, file = %s, line = %u\n",
498 (unsigned long long)bh->b_blocknr, type,
499 function, file, line);
500 me = gfs2_withdraw(sdp);
501 return (me) ? -1 : -2;
502}
503
504/*
505 * gfs2_metatype_check_ii - Flag a metadata type consistency error and withdraw
506 * Returns: -1 if this call withdrew the machine,
507 * -2 if it was already withdrawn
508 */
509
510int gfs2_metatype_check_ii(struct gfs2_sbd *sdp, struct buffer_head *bh,
511 u16 type, u16 t, const char *function,
512 char *file, unsigned int line)
513{
514 int me;
515
516 gfs2_lm(sdp,
517 fmt: "fatal: invalid metadata block\n"
518 " bh = %llu (type: exp=%u, found=%u)\n"
519 " function = %s, file = %s, line = %u\n",
520 (unsigned long long)bh->b_blocknr, type, t,
521 function, file, line);
522 me = gfs2_withdraw(sdp);
523 return (me) ? -1 : -2;
524}
525
526/*
527 * gfs2_io_error_i - Flag an I/O error and withdraw
528 * Returns: -1 if this call withdrew the machine,
529 * 0 if it was already withdrawn
530 */
531
532int gfs2_io_error_i(struct gfs2_sbd *sdp, const char *function, char *file,
533 unsigned int line)
534{
535 gfs2_lm(sdp,
536 fmt: "fatal: I/O error\n"
537 " function = %s, file = %s, line = %u\n",
538 function, file, line);
539 return gfs2_withdraw(sdp);
540}
541
542/*
543 * gfs2_io_error_bh_i - Flag a buffer I/O error
544 * @withdraw: withdraw the filesystem
545 */
546
547void gfs2_io_error_bh_i(struct gfs2_sbd *sdp, struct buffer_head *bh,
548 const char *function, char *file, unsigned int line,
549 bool withdraw)
550{
551 if (gfs2_withdrawing_or_withdrawn(sdp))
552 return;
553
554 fs_err(sdp, "fatal: I/O error\n"
555 " block = %llu\n"
556 " function = %s, file = %s, line = %u\n",
557 (unsigned long long)bh->b_blocknr, function, file, line);
558 if (withdraw)
559 gfs2_withdraw(sdp);
560}
561
562

source code of linux/fs/gfs2/util.c